Hyperspectral imaging and machine learning classify Medical Cannabis
University of the Basque Country project assists control of cannabis production.
Growing acceptance of medical cannabis has led to a worldwide increase in production of the plant, but some of the associated legal and regulatory framework has not kept pace.
According to a project at the University of the Basque Country (UPV/EHU), the current lack of quality control regulations or standards for correct manufacturing processes could lead to uncontrolled, and even harmful, crop products.
One possible answer could be enhanced optical imaging of the cannabis plants, and the UPV/EHU project has developed a technique based around hyperspectral imaging and machine learning designed to ensure the traceability or quality control of medicinal cannabis plants on an industrial scale.
Published in the journal Computers and Electronics in Agriculture, the study could pave the way for non-invasive horticultural quality control in medical cannabis production, serving an emerging industry that will require strict control over the cannabis chemotypes.
The new approach also has the advantage of avoiding destructive and time-consuming analytical techniques such as chromatography, noted the UPV/EHU team.
"Accurate, efficient methods need to be developed to ensure quality control in the plant production process," commented UPV/EHU's Markel San Nicolás. "Medical cannabis must be produced in a very controlled way and there is as yet no clear regulation in this regard."
Rapidly ensure traceability and quality control
Near-IR hyperpectral imaging (NIR-HSI) is attractive as a modality for this application since it enables an object to be visualized in two dimensions as a normal image, while also retrieving a wide electromagnetic spectrum from each pixel.
It also allows researchers to build on recent empirical research into the quantification of the main cannabinoid species of interest, THC and CBD, using NIR-HSI techniques, which indicated that the approach could simplify analysis of cannabis plants without compromising analytical capabilities.
In trials, NIR-HSI images of 57 cannabis plants were taken using a spectral range of 930 to 2500 nanometers divided into 288 spectral channels. Machine learning-based data analysis using trained algorithms then classified the target plants as belonging to chemotypes I, II or III, plant variants known to contain different amounts of cannabinoids.
The project's approach was able to successfully extract the three chemotypes of interest and determine the concentration of THC and CBD in the flower heads with an overall classification trueness of 94.74 percent, according to the team's paper.
A methodology based on NIR-HSI plus machine learning for chemotype classification can properly deal with the issue of moisture present in the fresh plant tissue, usually the main handicap when using conventional near-IR spectroscopy to analyze cannabinoids according to the project. The new approach also enables representative analysis directly in a complete living plant individual.
"Implementing this technology at medical cannabis production sites would automatically and rapidly ensure the traceability and quality control of the chemotype," commented Markel San Nicolás. "Although, for this to happen, cannabis-related regulation would have to be established and progress would have to be made in this industrial sector."
